Sparse Relational Topical Coding on multi-modal data

作者：

Highlights：

• A novel non-probabilistic relational topic model is proposed for modeling both multi-modal documents and the links between them.

• Sparse latent representations can be effectively learned through directly imposing appropriate regularizers.

• The proposed learning problem can be efficiently solved by a simple coordinate descent algorithm.

• The proposed model achieves significantly better performance than all the competing baseline models.

摘要

•A novel non-probabilistic relational topic model is proposed for modeling both multi-modal documents and the links between them.•Sparse latent representations can be effectively learned through directly imposing appropriate regularizers.•The proposed learning problem can be efficiently solved by a simple coordinate descent algorithm.•The proposed model achieves significantly better performance than all the competing baseline models.

论文关键词：Multi-modal data,Sparse latent representation,Image annotation,Link prediction

论文评审过程：Received 22 March 2017, Revised 25 July 2017, Accepted 3 August 2017, Available online 4 August 2017, Version of Record 17 August 2017.

论文官网地址：https://doi.org/10.1016/j.patcog.2017.08.005